YouSaid · the spoken record

Dileep George

lines on the record: 139
first: 2020-08-14
most recent: 2020-08-14
sittings or episodes: 1
sources: podcast

Every line below is reproduced as it was said and linked to the record it came from. Nothing here is summarised or generated. Directory · Search · Corrections

“So things like back propagation, credit assignment. So, many of these algorithms have learning algorithms, have things in common, right? It is a back propagation is one way of credit assignment. There is another algorithm called expectation maximization, which is another weight adjustment algorithm.”
2020-08-14 · Lex Fridman Podcast · #115 – Dileep George: Brain-Inspired AI · IDENTIFIED FROM THE TRANSCRIPT · source
“A lot of things are different, and those details matter a lot. So one point of difference I had with Jeff was how to approach how much of biological plausibility and realism do you want in the learning algorithms? When I was there, this was almost 10 years ago now. I don't know what just things now, but 10 years ago, the difference was that I did not want to be so constrained on saying my learning algorithms need to be biologically plausible based on some filter of biological plausibility available at that time. To me, that is a dangerous cut to make because we are discovering more and more things about the brain all the time. New biophysical mechanisms. Some new channels are being discovered all the time. So I don't want to upfront kill off a learning algorithm.”
2020-08-14 · Lex Fridman Podcast · #115 – Dileep George: Brain-Inspired AI · IDENTIFIED FROM THE TRANSCRIPT · source
“On intelligence. If you blur out the details and if you just zoom out, and at the higher level idea, things are, I would say, consistent with what he wrote about. But many things will be consistent with that because it's a blur, deep learning systems are also multi-level, hierarchical, all of those things. So in terms of the detail,”
2020-08-14 · Lex Fridman Podcast · #115 – Dileep George: Brain-Inspired AI · IDENTIFIED FROM THE TRANSCRIPT · source
“So, you need to understand the principles of processing for that. You can still apply engineering tricks. Where you want it to. You don't want to be slavishly mimicking all the things of the brain. So it should be one input. And I think it is extremely helpful. But it should be the point of really understanding so that you know when to deviate from it.”
2020-08-14 · Lex Fridman Podcast · #115 – Dileep George: Brain-Inspired AI · IDENTIFIED FROM THE TRANSCRIPT · source
“It has a phobia. And because of the phobia, the receptive fields are not like the copying of the weights. The weights in the center are very different from the weights in the periphery.”
2020-08-14 · Lex Fridman Podcast · #115 – Dileep George: Brain-Inspired AI · IDENTIFIED FROM THE TRANSCRIPT · source
“Yes, you need to. So I think it's one input and it is helpful, but you should know when to deviate from it too. So an example is convolutional neural networks. Convolution is not an operation brain implements. Visual cortex is not convolutional. Visual cortex has local receptive fields, local connectivity. But there is no translation in invariance in the network weights in the visual cortex. That is a computational trick, which is a very good engineering trick that we use for sharing the training between the different nodes. And that trick will be with us for some time. It will go away when we have Robots with eyes and heads that move. And so then that trick will go away. It will not be useful at that time.”
2020-08-14 · Lex Fridman Podcast · #115 – Dileep George: Brain-Inspired AI · IDENTIFIED FROM THE TRANSCRIPT · source
“My conference there is very high. I don't treat brain inspired as a marketing term. I am looking into the details of biology and puzzling over those things. And I am grappling with those things. So it is not a marketing term at all. You can use it as a marketing term. And people often use it. And you can get combined with them. And when people don't understand how we are approaching the problem, it is easy to be misunderstood and think of it as purely marketing. But that's not the way we are.”
2020-08-14 · Lex Fridman Podcast · #115 – Dileep George: Brain-Inspired AI · IDENTIFIED FROM THE TRANSCRIPT · source
“And so you can see where was what was being hyped in that thing, right? So it's like there is a dynamic in the community of that especially happens when there are lots of new people coming into the field and they get attracted to one thing and some people are trying to think different compared to that. So there is some, I think skepticism is science is important and it is very much required. But it's also not skepticism usually. It's mostly bandwagon effect that is happening rather than”
2020-08-14 · Lex Fridman Podcast · #115 – Dileep George: Brain-Inspired AI · IDENTIFIED FROM THE TRANSCRIPT · source
“I can tell you a story which is funny in the context of this, right? So if you read the abstract of the paper and the argument we are putting in, we are putting in, look, current deep learning systems take a lot of training data. They don't use these insights. And here is our new model, which is not a deep neural network, it's a graphical model. It does inference. This is how the paper is. Once the paper was accepted and everything, it went to the press department in science to play as science office. We didn't do any press release when it was published. We went to the press department. What was the press release that they wrote up? A new deep learning model.”
2020-08-14 · Lex Fridman Podcast · #115 – Dileep George: Brain-Inspired AI · IDENTIFIED FROM THE TRANSCRIPT · source
“Yeah, I can comment on it. So our Arsene paper is published in science, which I would argue is a very high quality journal, very hard to publish in. And usually it is indicative of the quality of the work. And I am very, very certain that the ideas that we brought together in that paper in terms of the importance of feedback connections, recursive inference, lateral connections, coming to best explanation of the scene as the problem to solve, trying to solve recognition, segmentation all jointly in a way that is compatible with higher level cognition, top-down attention, all those ideas that we brought together into something coherent and workable in the world and solving tackling a challenging problem, I think that will stay and that contribution I stand by, right?”
2020-08-14 · Lex Fridman Podcast · #115 – Dileep George: Brain-Inspired AI · IDENTIFIED FROM THE TRANSCRIPT · source
“Yeah, so the same model. So the important part of the model was that it trains very quickly with very little training data. And it's quite robust to out of distribution perturbations. And we are using that very fruitfully in advicarious in many of the robotic stuff we are solving.”
2020-08-14 · Lex Fridman Podcast · #115 – Dileep George: Brain-Inspired AI · IDENTIFIED FROM THE TRANSCRIPT · source
“That's correct, yeah. Because usually these things have a momentum once something gets established as a standard benchmark There is a dynamics of how graduate students operate and how academic system works that pushes people to track that benchmark.”
2020-08-14 · Lex Fridman Podcast · #115 – Dileep George: Brain-Inspired AI · IDENTIFIED FROM THE TRANSCRIPT · source
“Yeah, I remember that it was on the order of tens or hundreds of examples to get into 95% accuracy. And it was definitely better than the other systems out there at that time.”
2020-08-14 · Lex Fridman Podcast · #115 – Dileep George: Brain-Inspired AI · IDENTIFIED FROM THE TRANSCRIPT · source
“Yes, so we did do Amnest. So, you know, so it's not just CAPTHA. So there was also versions of multiple versions of Amnesty, including the standard version, which where we inverted the problem, which is basically saying rather than train on 60,000 training data, how quickly can you get to high level accuracy with very little training data?”
2020-08-14 · Lex Fridman Podcast · #115 – Dileep George: Brain-Inspired AI · IDENTIFIED FROM THE TRANSCRIPT · source
“And CAPCHA is a problem that is by definition hard for computers. And it has these good properties of strong generalization, strong out-of-training distribution generalization. If you are interested in studying that and having your model have that property, then it's a good data set to tackle.”
2020-08-14 · Lex Fridman Podcast · #115 – Dileep George: Brain-Inspired AI · IDENTIFIED FROM THE TRANSCRIPT · source
“Those benchmarks are useful for deep learning kind of algorithms where the settings that deep learning works in are here is my huge training set and here is my test set. So the training set is almost 100x, 1000x bigger than the test set in many cases. What we wanted to do was invert that. The training set is way smaller than the test set.”
2020-08-14 · Lex Fridman Podcast · #115 – Dileep George: Brain-Inspired AI · IDENTIFIED FROM THE TRANSCRIPT · source
“Not really hardcoded because it's the assumptions, as I mentioned, are general, right? It is more, and themselves can be applied in many situations which are natural signals. So it's the foreground versus background factory session and the fact-tray session of the surfaces versus the contours. So these are all generally applicable assumptions.”
2020-08-14 · Lex Fridman Podcast · #115 – Dileep George: Brain-Inspired AI · IDENTIFIED FROM THE TRANSCRIPT · source
“Then the kind of errors it makes are also, I don't want to read too much into it, but the kind of errors the network makes are very similar to the kinds of errors humans would make in a similar situation.”
2020-08-14 · Lex Fridman Podcast · #115 – Dileep George: Brain-Inspired AI · IDENTIFIED FROM THE TRANSCRIPT · source
“Yes, yeah. So, the good thing about the model is that it is extremely, so it is not just doing a classification, right? It is providing a full explanation for the scene. So when it operates on a scene, it is coming back and saying, look, this is the part is the A and these are the pixels that turned on the input that makes me think that it is an A. And also these are the portions I hallucinated. It provides a complete explanation of that form. And then these are the contours. This is the interior. And this is in front of this other object. So that's the kind of explanation the inference network provides. So that is useful and interpretable.”
2020-08-14 · Lex Fridman Podcast · #115 – Dileep George: Brain-Inspired AI · IDENTIFIED FROM THE TRANSCRIPT · source
“Dynamics is that even though locally it will look like, okay, this is an A. And locally, just when I look at just that batch of the image, it looks like an A. But when I look at it in the corner,”
2020-08-14 · Lex Fridman Podcast · #115 – Dileep George: Brain-Inspired AI · IDENTIFIED FROM THE TRANSCRIPT · source
“In this particular case, the temporal aspect is not important, it is more like if I turn the character on the pixels will turn on. It will be after a little bit, but.”
2020-08-14 · Lex Fridman Podcast · #115 – Dileep George: Brain-Inspired AI · IDENTIFIED FROM THE TRANSCRIPT · source
“So, you can think of it as you have at the top of the model the characters that you are trained on. Are the causes? You are trying to explain the pixels using the characters as the causes. The characters are the things that cause the pixels.”
2020-08-14 · Lex Fridman Podcast · #115 – Dileep George: Brain-Inspired AI · IDENTIFIED FROM THE TRANSCRIPT · source
“And just train the help out of that neural network, and it will look like it is doing inference on the fly, but it is really just doing amortized inference. Because you have shown it a lot of these combinations during training time. So what you want to do is be able to do dynamic inference rather than just being able to show all those combinations in the training time. And that's something we emphasized in the model.”
2020-08-14 · Lex Fridman Podcast · #115 – Dileep George: Brain-Inspired AI · IDENTIFIED FROM THE TRANSCRIPT · source
“And things like explaining away coming into this one. You are explaining that piece of evidence as something else because globally that's the only thing that makes sense. So now you can Amortize this inference by in a neural network. If you want to do this, you can brute force it. You can just show it all combinations of things that you want to your reasoning to work over.”
2020-08-14 · Lex Fridman Podcast · #115 – Dileep George: Brain-Inspired AI · IDENTIFIED FROM THE TRANSCRIPT · source
“So in captures what they do to confuse people is to make these characters crowd together. And when you make the characters crowd together, what happens is that you will now start seeing combinations of characters or some other new character or an existing character. So you would put an R and N together. It will start looking like an M. And so locally, there is very strong evidence for it being some incorrect character. But globally, the only explanation that fits together is something that is different from what you find locally. So, this is inference. You are basically taking local evidence and putting it in the global context and often coming to a conclusion locally, which is conflicting with the local information”
2020-08-14 · Lex Fridman Podcast · #115 – Dileep George: Brain-Inspired AI · IDENTIFIED FROM THE TRANSCRIPT · source
“Yeah. So, as I mentioned, one of the important things was being able to do inference, being able to dynamically do inference.”
2020-08-14 · Lex Fridman Podcast · #115 – Dileep George: Brain-Inspired AI · IDENTIFIED FROM THE TRANSCRIPT · source
“Essentially, yeah, right. So to finally solve, finally to say that you have solved capture, you have to solve the whole problem.”
2020-08-14 · Lex Fridman Podcast · #115 – Dileep George: Brain-Inspired AI · IDENTIFIED FROM THE TRANSCRIPT · source
“In all of its forms. It can be A can be made with two humans standing leaning against each other, holding the hands, and it can be made of leaves.”
2020-08-14 · Lex Fridman Podcast · #115 – Dileep George: Brain-Inspired AI · IDENTIFIED FROM THE TRANSCRIPT · source
“Yeah, you don't even know to go to the B and the C or the strings of characters. And so that is the spirit at which we tackle that problem.”
2020-08-14 · Lex Fridman Podcast · #115 – Dileep George: Brain-Inspired AI · IDENTIFIED FROM THE TRANSCRIPT · source
“Or no training examples from that particular style of capture. Even now, this is unreachable for the current deep learning system. So basically there is no, I don't think a system exists where you can basically say, train on whatever you want. And then now say, hey, I will show you a new capture, which I did not show you in the training setup. Will the system be able to solve it? Still doesn't exist. So that is the magic of human perception. And Doug Huffstarter put this very beautifully in one of his talks. The central problem in AI is what is the letter A? If you can build a system that reliably can detect all the variations of the letter A, you don't even need to go to the”
2020-08-14 · Lex Fridman Podcast · #115 – Dileep George: Brain-Inspired AI · IDENTIFIED FROM THE TRANSCRIPT · source
“Even now, I would say CAPCHA is a very, very good challenge problem if you want to understand how human perception works and if you want to build systems that work like the human brain. And I wouldn't say CAPCHA is a solved problem. We have cracked the fundamental defense of CAPCHAS, but it is not solved in the way that humans solve it. So I can give you an example. take a five-year-old child who has just learned characters and show them any new capture that we create. They will be able to solve it. I can show you pretty much any new capture from any new website. You'll be able to solve it without getting any training examples from that particular style of capture.”
2020-08-14 · Lex Fridman Podcast · #115 – Dileep George: Brain-Inspired AI · IDENTIFIED FROM THE TRANSCRIPT · source
“And text based captures was the one which is prevalent until around 2014 because at that time text-based waste captures were”
2020-08-14 · Lex Fridman Podcast · #115 – Dileep George: Brain-Inspired AI · IDENTIFIED FROM THE TRANSCRIPT · source
“So, CAPCHAs are those strings that you fill in, if you're opening a new account in Google, they show you a picture. Usually it used to be set of garbled letters that you have to kind of figure out what is that string of characters and type in. And the reason Capchast exists is because Google or Twitter do not want automatic creation of accounts. a computer to create millions of accounts and use that for in FADS purposes. So you want to make sure that to the extent possible, the interaction that their system is having is with a human. So it's called a human interaction proof. A capture is a human interaction proof. So this is a CAPCHA by design things that are easy for humans to solve but hard for computers.”
2020-08-14 · Lex Fridman Podcast · #115 – Dileep George: Brain-Inspired AI · IDENTIFIED FROM THE TRANSCRIPT · source
“And then there's this idea of doing inference. A neural network does not do inference on the fly. So an example of why this inference is important is one of the first applications that we showed in the paper was to crack text-based CAPCA”
2020-08-14 · Lex Fridman Podcast · #115 – Dileep George: Brain-Inspired AI · IDENTIFIED FROM THE TRANSCRIPT · source
“A constraint. It's basically if you do just feature detection followed by pooling, then your transformations in different parts of the visual field are not coordinated. And so you will create jagged when you generate from the model. You will create jagged things and uncoordinated transformations. So these lateral connections are enforcing the transformations.”
2020-08-14 · Lex Fridman Podcast · #115 – Dileep George: Brain-Inspired AI · IDENTIFIED FROM THE TRANSCRIPT · source
“Pathway. But in addition to that, it is also structured in a way that it is generative That it can run it backward and combine the forward with the backward. Another aspect that it has is it has lateral connections. This lateral connections. Which is between. So if you have an edge here and an edge here, it has connections between these edges. It is not just feed forward connections. It is something between these edges, which is the nodes representing these edges, which is to enforce compatibility between them. So otherwise what will happen is that...”
2020-08-14 · Lex Fridman Podcast · #115 – Dileep George: Brain-Inspired AI · IDENTIFIED FROM THE TRANSCRIPT · source
“Yeah, so you can think of the delta between the model and a convolutional neural network. If people are familiar with convolutional neural networks. So convolutional neural networks have this feed forward processing cascade, which is called feature detectors and pooling. And that is repeated in the hierarchy in a multi-level system. And intuitive area of what is happening, feature detectors are detecting interesting co-occurrences in the input. It can be a line, a corner, an eye or a piece of texture, etc. And the pulling neurons are doing some local transformation of that and making it invariant to local transformations. So this is what the structure of convolutional neuron network is. Recursive cortical network has a similar structure when you look at just the feed forward.”
2020-08-14 · Lex Fridman Podcast · #115 – Dileep George: Brain-Inspired AI · IDENTIFIED FROM THE TRANSCRIPT · source
“So, basically, saying, okay, these are the prior knowledge which will be derived from the work. But then how is that prior knowledge represented in the model such that inference when some piece of evidence comes in can be done very efficiently and in a very distributed way? Because there are so many ways of representing knowledge which is not amenable to very quick inference, quick lookups. So that's one core part of what we tackled in the RCN model. How do you encode visual knowledge to do very quick inference?”
2020-08-14 · Lex Fridman Podcast · #115 – Dileep George: Brain-Inspired AI · IDENTIFIED FROM THE TRANSCRIPT · source
“Drugged up rats out there. Okay, cool. From which these properties will emerge But it is still a very hard problem on how to encode that. So you don't, you know, there is no, so you mentioned the prior franchise wanted to encode in the abstract reasoning challenge, but it is not straightforward how to encode those priors. So some of those challenges, like the object completion challenges are things that we purely use our visual system to do. It looks like abstract reasoning, but it is purely an output of the vision system. For example, completing the corners of that Kaninza triangle, completing the lines of that Kanza triangle. It's a purely visual system property. There is no abstract reasoning involved. It uses all these priors, but it is stored in our visual system in a particular way that is amenable to influence. And that is one of the things that we tackled in the...”
2020-08-14 · Lex Fridman Podcast · #115 – Dileep George: Brain-Inspired AI · IDENTIFIED FROM THE TRANSCRIPT · source
“Yeah, but I think there are studies on that already. Already? Yeah, I think so. Because it's not unethical to give it to rats.”
2020-08-14 · Lex Fridman Podcast · #115 – Dileep George: Brain-Inspired AI · IDENTIFIED FROM THE TRANSCRIPT · source
“As opposed to a QR code, which is an artificial signal that we created, humans are not very good at classifying QR codes. We are very good at saying something is a cat or a dog, but not very good at where computers are very good at classifying QR codes. So our visual system is tuned for natural signals. And there are fundamental assumptions in the architecture that are derived from.”
2020-08-14 · Lex Fridman Podcast · #115 – Dileep George: Brain-Inspired AI · IDENTIFIED FROM THE TRANSCRIPT · source
“Keep in mind that you can derive this from much more general principles. You don't need to explicitly put it as objects versus foreground versus background, the surface versus texture. No, these are derivable from more fundamental principles of how what's the property of continuity of natural signals.”
2020-08-14 · Lex Fridman Podcast · #115 – Dileep George: Brain-Inspired AI · IDENTIFIED FROM THE TRANSCRIPT · source
“And then even that object is composed of parts. And also another one is the shape of the object is differently modeled from the texture of the object.”
2020-08-14 · Lex Fridman Podcast · #115 – Dileep George: Brain-Inspired AI · IDENTIFIED FROM THE TRANSCRIPT · source
“Quite a few things. It's like what does the model factorize? What is the model representing as different pieces in the puzzle? So in the RCN network, it thinks of the world. So the background of an image is modeled separately from the foreground of the image. So the objects are separate from the background. There are different entities.”
2020-08-14 · Lex Fridman Podcast · #115 – Dileep George: Brain-Inspired AI · IDENTIFIED FROM THE TRANSCRIPT · source
“You have no idea. No idea. All they have seen it millions of times. Hundreds of times. So it's not, our model is not photorealistic. But if you have other properties that we can manipulate it and you can think about filling in a different color in that logo, you can think about expanding the letter E. You can see, so you can imagine the consequence of actions that you have never performed. So these are the kind of characteristics the generative model need to have. So this is one constraint that went into our model. So this is, when you read just the perception side of the paper, it is not obvious that this was a constraint into that went into the model, this top-down controllability of the generating model.”
2020-08-14 · Lex Fridman Podcast · #115 – Dileep George: Brain-Inspired AI · IDENTIFIED FROM THE TRANSCRIPT · source
“Correct, correct. So basically having a generating network. Which is a model, and it is not just some arbitrary generated network, it has to be built in a way that it is controllable top-down. It is not just trying to generate a whole picture at once. It's not trying to generate photorealistic things of the world. You don't have good photorealistic models of the world. Human brains do not have, if I, for example, ask you the question, what is the color of the letter E in the Google logo?”
2020-08-14 · Lex Fridman Podcast · #115 – Dileep George: Brain-Inspired AI · IDENTIFIED FROM THE TRANSCRIPT · source
“So think of you can close your eyes and think about the details of one object. I can zoom in further and further. So think of the bottle in front of me. And now you can think about, okay, what the cap of that bottle looks. I know we can think about what's the texture on that bottle of the cap. You can think about what will happen if something hits that. So you can manipulate your visual knowledge in cognition-driven ways.”
2020-08-14 · Lex Fridman Podcast · #115 – Dileep George: Brain-Inspired AI · IDENTIFIED FROM THE TRANSCRIPT · source
“So there are multiple layers to the question. Again, go from the very top and then zoom in. So one important thing constraint that went into the model is that you should not think vision as something in isolation. We should not think perception as something as a pre-processor for cognition. Perception and cognition are interconnected. So you should not think of one problem in separation from the other problem. And so that means if you finally want to have a system that understand concepts about the world and can learn in a very conceptual model of the world and can reason and connect to language, all of those things, you need to think all the way through and make sure that your perception system is compatible with your cognition system and language system and all of them. And one aspect of that is top-down controllability.”
2020-08-14 · Lex Fridman Podcast · #115 – Dileep George: Brain-Inspired AI · IDENTIFIED FROM THE TRANSCRIPT · source
“On one side, it outputs the class of the image and also segments the image. And you can also ask it for further queries. Where is the edge of the object? Where is the interior of the object? So it's a model that you build to answer multiple questions. So you're not trying to build a model for just classification or just segmentation, et cetera. So it's a joint model that can do multiple things. So that's the model that we build using insights from neuroscience. And some of those insights are what is the role of feedback connections? What is the role of lateral connections? So all those things went into the model. The model actually uses feedback connections.”
2020-08-14 · Lex Fridman Podcast · #115 – Dileep George: Brain-Inspired AI · IDENTIFIED FROM THE TRANSCRIPT · source
“Okay, so let me step back a bit. So we looked at neuroscience for insights on how to build a vision model. And we synthesized all those insights into a computational model. This is called the recursive vertical network model that we.”
2020-08-14 · Lex Fridman Podcast · #115 – Dileep George: Brain-Inspired AI · IDENTIFIED FROM THE TRANSCRIPT · source